User repository on GitHub
A nano replica of Jev: parallel decisions, dynamic candidates, and an end-to-end training pipeline.
- Forks
- 263
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
NanoJev
An open 0.6B replica of Jev that turns states and questions into full probability distributions without decoding answer tokens, trained and evaluated on Maze, Snake, and ViZDoom.
- Stars
- 2.6k
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- TianyuCodings
- Last verified
- Sep 2026
About this project
A 0.6B parallel decision model: states and questions in, complete probability distributions out. Zero output-token decoding. Play ViZDoom · Maze & Snake · Model · Dataset. Now playing ViZDoom: one shared checkpoint handles Basic aiming and Predict Position's moving-target rocket shots, alongside Maze and Snake. 4 tasks · 18,760 decision questions per data variant · 896 Predict Position expert episodes. September 20, 2026 — One model, four games. ViZDoom Basic: 128/128 test successes, compared with 56/128 for Jev.
Quoted from the project's README. Read the full README.
Used inJev for Gaming
Highlights
- One shared step-400 checkpoint handles Maze, Snake, ViZDoom Basic, and Predict Position, with 18,760 questions per target data variant.
- Held-out test results: 128/128 on ViZDoom Basic versus 56/128 for Jev, and 27/128 on Predict Position versus 11/128.
- Choice returns a distribution over 2-255 candidates using set attention and softmax, Boolean uses a sigmoid, and Score returns a weighted level.
- Built on Qwen3-0.6B with decision heads and a persistent inference service; checkpoints and training data are published on Hugging Face.
Quickstart
bash
python -m pip install -r requirements-toy.txt huggingface_hubWatch out
MIT-licensed. Needs Python and a CUDA GPU to serve inference, and it is specialized to four game tasks rather than a general replacement for the Jev API.
Questions about this entry
Short answers drawn from the linked project or page.
On the held-out 274-case test set, with the same observation interface, candidate actions, and seeded controller: NanoJev takes 128/128 on ViZDoom Basic against Jev's 56/128, and 27/128 on Predict Position against 11/128. It goes the other way on Maze, 4/10 against Jev's 7/10, and ties 8/8 on Snake.
Not as a drop-in. One 0.6B backbone with decision heads is trained across Maze, Snake, ViZDoom Basic, and Predict Position, with 18,760 questions per target variant. It is a research replica for those four tasks, not a general replacement for the Jev API.
Between 2 and 255 supplied candidates, scored by set attention over candidate paths and a shared softmax. Booleans go through a sigmoid, and scores return a probability-weighted expectation over 2 to 10 ordered levels. Nothing is decoded as text.
The documented path serves from a local service in a CUDA environment, so plan on a CUDA GPU. Weights and the full dataset are public and need no sign-in; install is the toy requirements file plus huggingface_hub, and the service loads the model once.
Qwen3-0.6B with shared decision heads, reused across all four game tasks and a persistent inference service. The release tag unified-games-v1 packages the step-400 checkpoint from the hard_lr1e5 run.
Use huggingface_hub snapshot_download with revision unified-games-v1 for both the model repo C-Tianyu/NanoJev and the dataset repo C-Tianyu/NanoJev-Data. Both are public and need no sign-in.
18,760 per target variant, including 16,333 ViZDoom questions, with 10,898 in the training split. Of those, 10,893 pass the target-validity filter.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Show HN: OpenDecision, a local decision model with a Doom demo.
dwa35926 pointsShow HN: OpenDecision – a 400M zero-shot model makes local decisions, plays DoomRead the thread on Hacker NewsX: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsMore Jev resources like this
An open, non-autoregressive System One model with published weights and a TypeSafe-compatible /v1/systemone API; the 395M-parameter OptionMarker checkpoint scores discrete and ordinal questions locally.
Jev Repos & SDKs#community#python#open-models
Community
wfzyx/vonPython
wfzyx/vonPythonTurns local text and vision models into Jev-style decision models by reading probabilities from prefill logits alone, with SGLang, Transformers, and Apple Silicon MLX backends.
Jev Repos & SDKs#community#python#open-models
Community
Yinsongxu/LLM2JevPython
Yinsongxu/LLM2JevPythonAn educational Jev-like inference project for Apple Silicon: Qwen3.5-0.8B scores candidate answers from shared multimodal context instead of generating text, with browser, CLI, and HTTP interfaces.
Jev Repos & SDKs#community#python#open-models
Community
hr98w/jev-visualPython
hr98w/jev-visualPythonAn open 0.6B System One decision model that reads unstructured state and typed questions in one forward pass, returning calibrated distributions with zero decoded tokens; weights are published on Hugging Face.
Jev Repos & SDKs#community#python#open-models
Community
malevrigns/agent-jevPython
malevrigns/agent-jevPythonAn Apache-licensed Python library that reads option-token probabilities from open-weight models through Transformers or vLLM, giving Jev-style typed decisions in one forward pass without an HTTP server.
Jev Repos & SDKs#community#python#open-models
Community
ikermoel/open-alternative-jevPython
ikermoel/open-alternative-jevPythonA reproduction of Jev that turns any Qwen checkpoint into a decision model serving the same /v1/systemone schema (Choice, Score, Noul), with no training and no generated answer text.
Jev Repos & SDKs#community#python#open-models
Community
zhengxuyu/litjevPython
zhengxuyu/litjevPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.